CN1241149C - Compression rate pre-allocation algorithm for JPEG 2000 multi-picture photo - Google Patents

Abstract

The present invention discloses a preallocation algorithm of the compression ratio of JPEG2000 images. The preallocation algorithm of the present invention can effectively allocate compression ratios set by users to images according to the information amount contained of the images in JPEG2000 processing, and therefore, solves the problems of poor chrominance and rough edges generated by that the whole compression ratios are simply and averagely allocated to each image. A JPEG-LS forecast moulding board, a PNG forecast moulding board and an edge detection operator are introduced for being used for calculating the information content of the images. Through the indication of a large number of experimental results, the visual effects of reconstructed images can obtain different degree improvement and reinforcement through the use of the preallocation algorithm of the compression ratio of JPEG2000 images, mean square errors (MSE)are remarkably reduced, and the preallocation algorithm is especially effective on images with targets with area concentricity. The preallocation algorithm of the present invention has good real time performance and good effectiveness by using a JPEG-LS prediction algorithm, a PNG prediction algorithm and edge detection operators of a small moulding board to measure the information amount of the images, and can satisfy the requirements of hardware.

Description

A kind of many image sheets of JPEG2000 compressibility method for pre-distributing

One, technical field

The invention belongs to the VLSI design field.Be specifically related in JPEG2000 hardware is realized, design a kind of new JPEG2000 compressibility method for pre-distributing.

Two, background technology

JPEG2000 still image compression standard has become the focus that engineering circle and academia are paid close attention to always since its issue.All being to carry out at an image sheet all about compressibility control (rate control) part in the JPEG2000 agreement, how it should distribute time-limited code word if at large having illustrated in an image sheet, just can make the distortion behind the image reconstruction reach minimum.But, for bigger image, generally all can be divided into a plurality of image sheets to it and handle, all forward direction component transformations, DC displacement, wavelet transform, quantification, arithmetic coding and layering are blocked and are packed, and all only can carry out in the inside of an image sheet.Does should how to carry out compressibility so between image sheet distribute? problem hereto, from all JPEG2000 agreements of having issued on present recognize internal and international with about the paper of JPEG2000 compressibility control aspect and in view of the JPEG2000 software and hardware structure is realized, all not have this problem of relevant documents, the solution that does not more have proposition to be correlated with.From the JasPer/JPEG2000 source program, it does not consider the information difference between the image sheet, just gives each image sheet with the reduced overall rate mean allocation of image.The applicant finds through observational study, and when information distributes relatively more even on entire image or ratio of compression relatively low the time, this disposal route is also barely satisfactory; If but information is under the situation of extremely uneven and high compression ratio that distributes between each image sheet, with the method compressed image can the tangible colour difference of appearance between image sheet and sheet and the edge rough.Certainly can not allow the user manually set the compressibility of each image sheet, should have a cover to carry out the automatic adaptive algorithm of distributing of compressibility and go to address this problem according to the image sheet information content.

But the evaluation method of traditional quantity of information will be added up the pixel value of image as the Shannon quantity of information, but also will carry out logarithm operation.Same MSE statistical method need be carried out twice scanning to original image, but also will be carried out square operation.Therefore the computing method of traditional information content can't satisfy the real-time requirement of hardware because its calculated amount is quite big, and it is necessary therefore finding a kind of appropriate information amount statistical method to come the information content of evaluation image.

Three, summary of the invention

Defective and deficiency according to the above-mentioned background technology exists the objective of the invention is to, and a kind of many image sheets of JPEG2000 compressibility method for pre-distributing is provided.

The solution that the present invention adopts is: the information content of computed image at first, between image sheet, carry out the predistribution of compressibility then according to the information content of image; Introduce the prediction module of JPEG-LS and PNG and the information content that the rim detection module is weighed image in addition;

At least may further comprise the steps:

1) introduces prediction module and edge detection operator and carry out the quantity of information evaluation of image sheet

A) adopt JPEG-LS or PNG prediction module, predict in the scope of neighbours territory, establish x ' and be premeasuring, then information of forecasting content is: x-x ' (x is the actual pixels value), and the information content of overall image graphic web is the summation of the information of forecasting content of all pixels;

B) adopt edge detection operator to estimate the method for the quantity of information of image: at first to act on former image, obtain the quantity of information of each pixel with edge detection template; Each Pixel Information amount summation is exactly an information content of image;

2) product of the ratio of compression of setting according to the information content of different images sheet and user distributes the ratio of compression of each image sheet as weighting coefficient.

Many image sheets of JPEG2000 compressibility predistribution algorithm of the present invention, the visual effect of reconstructed image can obtain improvement and reinforcement in various degree, and MSE significantly reduces, and is for the image that target has set of regions neutrality, particularly effective; The quantity of information of weighing image sheet with the edge detection operator of JPEG-LS and PNG prediction algorithm and little template has good real-time performance and validity, can satisfy the requirement of hardware design.

Four, description of drawings

Fig. 1 is a forecasting process synoptic diagram of the present invention;

Fig. 2 is a prediction module synoptic diagram of the present invention;

Fig. 3 is a JPEG-LS prediction algorithm synoptic diagram;

Fig. 4 is a PNG prediction algorithm synoptic diagram;

Fig. 5 is the edge detection process synoptic diagram;

Fig. 6 is the edge detection template synoptic diagram of simplifying;

Fig. 7 is the JPEG2000 coding process flow diagram after improving.

Five, embodiment

The present invention carries out the method that compressibility is distributed automatically according to the image sheet information content between image sheet.The present invention is described in further detail below in conjunction with accompanying drawing.

1. the evaluation method of information content

The evaluation method of image information content has a lot, uses the Shannon quantity of information to weigh usually.But the Shannon quantity of information need be added up the pixel value of being had a few, but also will carry out logarithm operation to the statistical probability that obtains, and calculated amount is very big, and real-time is poor.Same square error (MSE) MSE also is a kind of method of weighing image information content, and it need carry out twice scanning to entire image, but also will carry out square operation, so calculated amount is also quite big, the more important thing is that it can't realize concurrent operation.In addition, Shannon quantity of information and MSE statistical method are not all considered the realm information of image pixel.At the defective of top two kinds of methods, introduce the practical algorithm that rim detection and prediction module are weighed image information content first, the real-time of these two operators and validity are better than Shannon and MSE greatly.

* information of forecasting content adds a prediction module to image, carries out difference with predicted value and source pixel value then, carries out the calculating of information content at last with this difference.

The calculation process of information of forecasting content as shown in Figure 1, this prediction module is based on mainly that four fields (Fig. 2) predict, common prediction module has JPEG-LS (Fig. 3) and PNG (Fig. 4), these two prediction module have the characteristics of three-favour:

1) calculated amount is little;

2) considered realm information;

3) more meet the definition of quantity of information.

* based on the statistic law of rim detection

The edge is very important characteristic concerning piece image, so the edge also is an important symbol weighing image information content.Introducing some edge detection templates and detect edge of image earlier, determine the information content of image again with the edge, also is a kind of important method.

Common edge detection operator has Roberts crossover operator, Sobel operator, Prewitt operator, Laplace operator, LOG operator and Canny operator etc.The edge is abundant more, and the information content that image comprised is just big more.The first-elected Canny operator of optimum operator at check edge, the edge that it not only can detected image, and have the function of denoising.Found through experiments, after use 9 * 9Canny operator carried out the information content statistics, the average effect that compressibility is distributed was best really, and stability is high.Blemish in an otherwise perfect thing is that operand is too huge, no matter is hardware or software, and probably its expense all is difficult to bear.

The present invention recommends a kind of edge detection operator (as Fig. 6) of simplification, uses this template not need to carry out multiplying, and efficient improves greatly.In practicality, can also carry out interval sampling by per 4 points or 8 points, and then carry out rim detection.So not only improve arithmetic speed, and can detect big edge, be not vulnerable to interference of noise.Adopting this method only need pay very little computing cost, just can significantly improve picture quality under the constant situation of ratio of compression, is a kind of compressibility method for pre-distributing of keypoint recommendation of the present invention.

2. pre-allocation process

Make that ImageSize is an original image size, N is the image sheet number that marks off, and ratio is the given ratio of compression of user, the compressibility weight coefficient ri of each image sheet then, and i=1,2,3 ... N, satisfy (2.1) formula:

$\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\mathrm{ri}=1---\left(2.1\right)$

The assigned byte number of each image sheet is R _i, satisfy (2.2) formula:

(2.2)

Ri＝ImageSize×ratio×ri

Compressibility weight coefficient r _kSatisfy (2.3) formula:

$r_k=\frac{I_k}{\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{I}_i}---(2.3)$

I wherein _kBe the pairing information content of image sheet k, can obtain with any one given evaluation method of front.Add compressibility predistribution module JPEG2000 coding flow process afterwards as shown in Figure 7.

Use is tested many images with set of regions neutrality based on the predistribution algorithm of image sheet information content, and the result shows: the effect behind the Compress softwares can significantly be strengthened, and this algorithm has higher real-time and validity.

The information content of image sheet refers to the needed minimal compression storage space of restructural piece image sheet.The definite quantitative criteria of a kind of meaning neither one in back, it is closely related with the algorithm of concrete removal redundant information.Therefore for image, not just relevant with the pixel value of image, and arrange closely related with locations of pixels.And Shannon quantity of information and MSE statistic are not considered locations of pixels information.They have considered all and the relation of realm information that therefore this information content is not only the statistics of having expressed pixel value and is the more important thing is the accumulative total that has comprised location of pixels information prediction module and rim detection module.The calculated amount of prediction module and edge detection operator is little in addition, only need a front end in the image input add the row of 2～3 images deposit can real-time calculating chart photo quantity of information, and need not repeatedly scan image.

Compressibility predistribution algorithm is meant the size according to the information content of each image sheet, and the compressibility that the user is set is assigned to the method in each image sheet.At first calculate the information content of each image sheet according to the method for information evaluation, and then according to the difference of quantity of information, distribute different compressibility right of distribution coefficients for each image sheet, and then multiply each other with compressibility that the user sets and just can obtain the realistic compression ratio of present image sheet.

A large amount of test findings show: use compressibility predistribution algorithm, the visual effect of reconstructed image can obtain improvement and reinforcement in various degree, and MSE significantly reduces, and is for the image that target has set of regions neutrality, particularly effective; The quantity of information of weighing image sheet with the edge detection operator of JPEG-LS and PNG prediction algorithm and little template has good real-time performance and validity, can satisfy the requirement of hardware design.

Claims (2)

1. many image sheets of JPEG2000 compressibility method for pre-distributing is characterized in that, at first the information content of computed image is carried out the predistribution of compressibility then between image sheet according to the information content of image; Introduce the prediction module of JPEG-LS and PNG and the information content that the rim detection module is weighed image in addition;

At least may further comprise the steps:

1) introduces prediction module and edge detection operator and carry out the quantity of information evaluation of image sheet;

A) adopt JPEG-LS or PNG prediction module, predict in the scope of neighbours territory, establish x ' and be premeasuring, x is the actual pixels value, and then information of forecasting content is: x-x ', and the information content of overall image graphic web is the summation of the information of forecasting content of all pixels;

B) adopt the method for the quantity of information of edge detection operator evaluation image to be: at first to act on former image, obtain the quantity of information of each pixel with edge detection template; Each Pixel Information amount summation is exactly an information content of image;

2) product of the ratio of compression of setting according to the information content of different images sheet and user distributes the ratio of compression of each image sheet as weighting coefficient.

2. many image sheets of JPEG2000 compressibility method for pre-distributing as claimed in claim 1 is characterized in that described method for pre-distributing carries out as follows:

Make that ImageSize is an original image size, N is the image sheet number that marks off, and ratio is the given ratio of compression of user, and then the compressibility weight coefficient of each image sheet is ri, i=1, and 2,3 ... N, satisfy (2.1) formula:

$\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\mathrm{ri}=1---\left(2.1\right)$

The assigned byte number of each image sheet is R _i, satisfy (2.2) formula:

(2.2)

Ri＝ImageSize×ratio×ri

Compressibility weight coefficient r _kSatisfy (2.3) formula;

$r_k=\frac{I_k}{\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{I}_i}------\left(2.3\right)$

I wherein _kBe the pairing information content of image sheet k.

Images